Hybrid Force-Position Robot Control: An Artificial Neural Network Backstepping Approach

Abstract

We derive an adaptive Lyapunov backstepping scheme to achieve hybrid force-position control of a revolute-joint robotic manipulator. It is suitable for the situation where the desired force and desired trajectory motion are perpendicular i.e. for operating on a flat surface. The control also tracks commands in free space so that no switching is required when encountering/leaving the surface. The control utilizes the robot parameters but neural networks adaptively model the environmental effects. The proof of stability requires an assumption of a passive mapping from velocity to force and that the environment can be modelled as a nonlinear stiffness. Simulation results show the proposed neural-adaptive solution can, without any pretraining, significantly outperform linear methods in both position and force tracking.